agree with the two things but I would add a third ..
tell me what happens when your server is unreachable.. not just what it does when it works.
most of the mcp writeups I have read cover the happy path in detail and the failure path in one line and thats backwards for anything going near production.
what does your server actually do when the call times out?
price per token on its own tells you nothing.. what matters is tokens per successful outcome a cheap model that needs three retries costs more than an expensive one that gets it right first go.
anyone actually tracking cost per completed task rather than cost per call ? it is quite a difficult task to define completion as it is a qualitative aspect and can vary from every individual.
the meaningless part is right but doesnt go far enough.. price per token also assumes every token does the same amount of work..
a model that reasons for 10x longer to get the same answer is not actually cheaper just because its rate is lower.
the number that actually tells you something is cost to get a task done end to end.. not the sticker price on a token. anyone tracking that instead of the headline rate ??
Everyone here is arguing about what the agent could read but the leak only happened because it could write the data back out as a public comment on the issue. That is the half worth cutting.. You will never win the injection fight on the input side but an agent triggered by a public issue shouldn't be able to post public output containing anything it pulled from a private scope. The scary sounding permission is the read .. the one that actually leaked is the public write back.
The actual damage comes from volume without accountability.. no attribution, no source no way to verify. The same problem existed with content farms pre AI it's just cheaper and rampant now. Communities that survived content farms did it with one rule.. cite your source or get removed. That still works. but again the irony is that the source could be some ai slop..
OpenAI API has had 7 incident-days this month. Anthropic has been relatively clean outside this one. The frustrating part is you usually find out from Twitter before the status page updates.